market-inflection-radar

name

market-inflection-radar

description

Strict post-close industrial investment inflection scan for rare S-level opportunities. Use when the user asks to run today's market-inflection-radar, run a post-close产业级投资机会拐点扫描, evaluate S-level industrial opportunity candidates, run historical run-date scans, deep-dive a theme, or execute golden-test regression cases. The workflow distinguishes first-confirmed industry expectation-curve upgrades from daily stock movement explanations and defaults to no S-level opportunity unless every gate passes.

Market Inflection Radar

Use this skill as an ultra-low-frequency radar for "产业级预期曲线首次上修并被市场确认" opportunities. Do not use it as a daily stock picker, news explainer, theme heat map, or forum narrative summarizer.

Non-Negotiable Start

When triggered by a one-line request such as "运行今日 market-inflection-radar", run the CLI before writing any conclusion. Do not produce a report from intuition.

From this skill directory:

cd scripts
python -m inflection_radar run-today --adapter live
python -m inflection_radar run-date --date YYYY-MM-DD --adapter live
python -m inflection_radar deep-dive --theme xxx --date YYYY-MM-DD
python -m inflection_radar sources list --adapter live
python -m inflection_radar sources query --adapter live --source market --date YYYY-MM-DD --entity MU
python -m inflection_radar collect --adapter live --date YYYY-MM-DD --theme xxx --out /tmp/mir-source-bundle.md
python -m inflection_radar golden-test
python -m inflection_radar blind-packet --case all --output-dir /tmp/mir-blind

If running from another working directory, set PYTHONPATH to this skill's scripts/ directory before the same commands.

If the local shell does not expose python, use python3 -m inflection_radar ... with the same arguments.

Default Stance

Start from: 今日无 S 级机会.

Only override that default if at least one candidate passes all seven gates:

1. Industry space is huge.
2. The thesis is hard to disprove in the short term.
3. A real external catalyst changed expectations.
4. The market printed a first or early large bullish confirmation.
5. There are high-purity beneficiary assets.
6. Odds are still reasonable versus price location and crowding.
7. Bear-case review does not break the thesis.

If any gate fails, do not output S. Report the first failed gate and cap the candidate at A/B/C according to references/scoring-rubric.md.

Operating Workflow

1. Run run-today --adapter live for a daily request. This command must scan the previous trading day using the built-in daily theme watchlist; do not stop merely because no pre-existing candidate was supplied.
2. Treat the CLI output as an automated source sweep, not a full conclusion. It must show theme coverage, source counts, and search-required follow-ups.
3. For production-like scans, run sources list --adapter live to confirm public adapters, then run targeted collect --adapter live --date ... --theme ... for any theme that looks non-empty or gate-relevant. Use fixture only for tests.
4. Read references/mission.md and references/opportunity-definition.md if the task is ambiguous.
5. For any possible S candidate, load references/scoring-rubric.md, references/evidence-standards.md, references/anti-hype-checklist.md, and references/report-template.md.
6. Build an evidence table before writing the rating. High-grade evidence must anchor the core thesis; low-grade evidence may only describe narrative spread or crowding.
7. If a live source returns search-required, unavailable, or zero records for a gate-critical claim, use available web/search/browser tools to find official or high-grade sources with the scan date as a hard cutoff. Log the query, source, date, and whether it failed. Do not treat a missing source as verified.
8. Apply the anti-hype checklist after the positive case, not before it. A single hard failure there blocks S.
9. Use references/golden-cases.md as evaluator-only regression behavior. Do not feed its answer-key language to an agent being forward-tested.
10. For blind validation, generate packets with python -m inflection_radar blind-packet --case all --output-dir /tmp/mir-blind and give the target agent only the generated packet plus this Skill. The packet must contain only dated, as-of information, not expected ratings or correct labels. The source packet files live in references/golden-packets/.

Output Rules

For no S-level setup, keep the answer short:

一句话结论：今日无 S 级机会。
主要观察：...
被否候选：候选 / 最高评级 / 首个未通过 gate / 需要继续跟踪的证据。

For an S-level setup, use references/report-template.md exactly as the report skeleton. Include:

• 一句话结论
• 今日触发信号
• 为什么不是普通异动
• 产业空间
• 受益链条
• 核心标的分层
• 证据表
• 反方观点
• 证伪条件
• 拥挤度与时点
• 最终评级

Always phrase the output as research triage, not investment advice or a buy/sell instruction.

References

• references/mission.md: mission, non-goals, and expected behavior.
• references/opportunity-definition.md: what counts as an S-level industrial inflection.
• references/trigger-taxonomy.md: catalyst taxonomy and trigger patterns.
• references/scoring-rubric.md: hard gate system and rating caps.
• references/evidence-standards.md: source hierarchy and evidence rules.
• references/report-template.md: mandatory S-level report format.
• references/anti-hype-checklist.md: hard blocks against hype-driven S ratings.
• references/golden-cases.md: evaluator-only regression answer key and blind-test protocol.
• references/golden-packets/: blind historical packets for forward testing without answer keys.
• references/theme-ontology.schema.md: theme/entity schema for structured candidates.
• references/data-sources.md: preferred data source categories and collection order.